Demis Hassabis Endorses Jensen Huang's Open-Weight AI Push: How Silicon Valley's Safety vs. Openness Divide Is Crumbling

Core Arguments of the Letter

• Open-weight models strengthen safety and cybersecurity, accelerate innovation, and enable AI sovereignty for other nations .
• Limiting open models would reduce competition and drive more users toward Chinese AI alternatives .
• The world needs a "plural frontier"—both closed models (like GPT-4o or Gemini) and open models (like Llama or Mistral) should coexist .
• Premature regulation would stifle American leadership in AI at a moment when Chinese models like Moonshot AI's Kimi K3 are advancing rapidly .

The Signatories: Who Signed and Who Didn't

The original letter, released on July 24, had 25 signatories including Nvidia, Microsoft, Meta, IBM, Dell Technologies, Palantir, Hugging Face, Mistral, Mozilla, Y Combinator, Andreessen Horowitz, Replit, Perplexity, and others . Notably, OpenAI, Google (Alphabet), Amazon, and Anthropic were initially absent .

Within a day, after public attention on X, the list expanded to roughly 50 signatories. New joiners included OpenAI, Google (Alphabet), AMD, Cisco, Cloudflare, GitHub, Block, and Ollama. Amazon and Anthropic remained non-signatories .

What This Reflects: A Shifting Silicon Valley Consensus

Hassabis's endorsement, alongside the rapid expansion of the signatory list, reflects a meaningful shift in the industry's center of gravity:

• The old divide pitted "safety-first" labs (DeepMind, Anthropic, OpenAI until recently) against "open-source advocates" (Meta, Mistral, Hugging Face). The two camps were often portrayed as incompatible.
• The emerging consensus—visibly backed by Hassabis, a leading safety researcher—holds that openness and safety are not a binary choice. The letter itself argues that open models contribute to safety by enabling external auditing, red-teaming, and decentralized security research .
• Commercial self-interest aligns: Nvidia benefits from broader model adoption regardless of openness; Meta and Mistral have open-weight business models; and Microsoft and IBM hedge across both approaches. Even safety-focused labs now see that a total ban on open weights would hand AI leadership to China .
• The notable holdouts—Amazon and especially Anthropic—show the consensus is not universal. Anthropic continues to argue that unrestricted open weights carry unacceptable catastrophic risks, and its absence kept that fault line visible .

The Bottom Line

Hassabis's public backing turned a corporate lobbying letter into a visible signal that the 'openness vs. safety' framing is breaking down inside Silicon Valley's most cautious AI labs, even as real disagreements persist on the margins. The coalition now includes roughly 50 companies spanning chipmakers, cloud providers, venture investors, and open-model labs—a breadth that would have been unthinkable just two years ago.